edu.cmu.sphinx.decoder.search

Class SimpleBreadthFirstSearchManager

java.lang.Object

edu.cmu.sphinx.decoder.search.TokenSearchManager

edu.cmu.sphinx.decoder.search.SimpleBreadthFirstSearchManager

All Implemented Interfaces:

SearchManager, Configurable

public class SimpleBreadthFirstSearchManager
extends TokenSearchManager

Provides the breadth first search. To perform recognition an application should call initialize before recognition begins, and repeatedly call recognize until Result.isFinal() returns true. Once a final result has been obtained, terminate should be called.

All scores and probabilities are maintained in the log math log domain.

For information about breadth first search please refer to "Spoken Language Processing", X. Huang, PTR

Field Summary

Fields

Modifier and Type	Field and Description
protected ActiveList	activeList
protected ActiveListFactory	activeListFactory
protected java.util.Map<SearchState,Token>	bestTokenMap
protected long	currentCollectTime
protected int	currentFrameNumber
protected Timer	growTimer
protected Linguist	linguist
protected LogMath	logMath
static java.lang.String	PROP_ACTIVE_LIST_FACTORY The property that defines the name of the active list factory to be used by this search manager.
static java.lang.String	PROP_GROW_SKIP_INTERVAL The property that controls the number of frames processed for every time the decode growth step is skipped.
static java.lang.String	PROP_LINGUIST The property that defines the name of the linguist to be used by this search manager.
static java.lang.String	PROP_PRUNER The property that defines the name of the linguist to be used by this search manager.
static java.lang.String	PROP_RELATIVE_WORD_BEAM_WIDTH The property that sets the minimum score relative to the maximum score in the word list for pruning.
static java.lang.String	PROP_SCORER The property that defines the name of the scorer to be used by this search manager.
static java.lang.String	PROP_SHOW_TOKEN_COUNT The property that when set to true will cause the recognizer to count up all the tokens in the active list after every frame.
static java.lang.String	PROP_WANT_ENTRY_PRUNING The property that controls whether or not relative beam pruning will be performed on the entry into a state.
protected java.util.List<Token>	resultList
protected boolean	showTokenCount
protected boolean	streamEnd

Fields inherited from class edu.cmu.sphinx.decoder.search.TokenSearchManager

buildWordLattice, keepAllTokens, PROP_BUILD_WORD_LATTICE, PROP_KEEP_ALL_TOKENS

Constructor Summary

Constructors

Constructor and Description
SimpleBreadthFirstSearchManager()
SimpleBreadthFirstSearchManager(Linguist linguist, Pruner pruner, AcousticScorer scorer, ActiveListFactory activeListFactory, boolean showTokenCount, double relativeWordBeamWidth, int growSkipInterval, boolean wantEntryPruning) Creates a manager for simple search

Method Summary

Modifier and Type	Method and Description
void	allocate() Allocates the resources necessary for this search.
protected void	collectSuccessorTokens(Token token) Collects the next set of emitting tokens from a token and accumulates them in the active or result lists
void	deallocate() Deallocates resources necessary for this search.
ActiveList	getActiveList()
protected Token	getBestToken(SearchState state) Gets the best token for this state
protected java.util.Map<SearchState,Token>	getBestTokenMap() Returns the best token map.
int	getCurrentFrameNumber() Returns the current frame number.
Timer	getGrowTimer() Returns the Timer for growing.
java.util.List<Token>	getResultList() Returns the result list.
StatisticsVariable	getTokensCreated() Returns the tokensCreated StatisticsVariable.
protected void	growBranches() Goes through the active list of tokens and expands each token, finding the set of successor tokens until all the successor tokens are emitting tokens.
protected void	localStart() Gets the initial grammar node from the linguist and creates a GrammarNodeToken
protected void	localStop() Local cleanup for this search manager
void	newProperties(PropertySheet ps) This method is called when this configurable component needs to be reconfigured.
protected void	pruneBranches() Removes unpromising branches from the active list
protected boolean	recognize() Performs recognition for one frame.
Result	recognize(int nFrames) Performs the recognition for the given number of frames.
protected boolean	scoreTokens() Calculate the acoustic scores for the active list.
protected Token	setBestToken(Token token, SearchState state) Sets the best token for a given state
protected void	setBestTokenMap(java.util.Map<SearchState,Token> bestTokenMap) Sets the best token Map.
protected void	showTokenCount() Counts all the tokens in the active list (and displays them).
void	startRecognition() Called at the start of recognition.
void	stopRecognition() Terminates a recognition
java.lang.String	toString()
protected ActiveList	undoLastGrowStep() Because the growBranches() is called although no data is left after the last speech frame, the ordering of the active-list might depend on the transition probabilities and (penalty-scores) only.

Methods inherited from class edu.cmu.sphinx.decoder.search.TokenSearchManager

getResultListPredecessor

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Detail

PROP_LINGUIST

@S4Component(type=Linguist.class)
public static final java.lang.String PROP_LINGUIST

The property that defines the name of the linguist to be used by this search manager.

See Also:

Constant Field Values

PROP_PRUNER

@S4Component(type=Pruner.class)
public static final java.lang.String PROP_PRUNER

The property that defines the name of the linguist to be used by this search manager.

See Also:

Constant Field Values

PROP_SCORER

@S4Component(type=AcousticScorer.class)
public static final java.lang.String PROP_SCORER

The property that defines the name of the scorer to be used by this search manager.

See Also:

Constant Field Values

PROP_ACTIVE_LIST_FACTORY

@S4Component(type=ActiveListFactory.class)
public static final java.lang.String PROP_ACTIVE_LIST_FACTORY

The property that defines the name of the active list factory to be used by this search manager.

See Also:

Constant Field Values

PROP_SHOW_TOKEN_COUNT

@S4Boolean(defaultValue=false)
public static final java.lang.String PROP_SHOW_TOKEN_COUNT

The property that when set to true will cause the recognizer to count up all the tokens in the active list after every frame.

See Also:

Constant Field Values

PROP_RELATIVE_WORD_BEAM_WIDTH

@S4Double(defaultValue=0.0)
public static final java.lang.String PROP_RELATIVE_WORD_BEAM_WIDTH

The property that sets the minimum score relative to the maximum score in the word list for pruning. Words with a score less than relativeBeamWidth * maximumScore will be pruned from the list

See Also:

Constant Field Values

PROP_WANT_ENTRY_PRUNING

@S4Boolean(defaultValue=false)
public static final java.lang.String PROP_WANT_ENTRY_PRUNING

The property that controls whether or not relative beam pruning will be performed on the entry into a state.

See Also:

Constant Field Values

PROP_GROW_SKIP_INTERVAL

@S4Integer(defaultValue=0)
public static final java.lang.String PROP_GROW_SKIP_INTERVAL

The property that controls the number of frames processed for every time the decode growth step is skipped. Setting this property to zero disables grow skipping. Setting this number to a small integer will increase the speed of the decoder but will also decrease its accuracy. The higher the number, the less often the grow code is skipped.

See Also:

Constant Field Values

linguist

protected Linguist linguist

currentFrameNumber

protected int currentFrameNumber

currentCollectTime

protected long currentCollectTime

activeList

protected ActiveList activeList

resultList

protected java.util.List<Token> resultList

logMath

protected LogMath logMath

growTimer

protected Timer growTimer

showTokenCount

protected boolean showTokenCount

bestTokenMap

protected java.util.Map<SearchState,Token> bestTokenMap

activeListFactory

protected ActiveListFactory activeListFactory

streamEnd

protected boolean streamEnd

Constructor Detail

SimpleBreadthFirstSearchManager

public SimpleBreadthFirstSearchManager()

SimpleBreadthFirstSearchManager

public SimpleBreadthFirstSearchManager(Linguist linguist,
                                       Pruner pruner,
                                       AcousticScorer scorer,
                                       ActiveListFactory activeListFactory,
                                       boolean showTokenCount,
                                       double relativeWordBeamWidth,
                                       int growSkipInterval,
                                       boolean wantEntryPruning)

Creates a manager for simple search

Parameters:

linguist - linguist to configure search space

pruner - pruner to prune extra paths

scorer - scorer to estimate token probability

activeListFactory - factory for list of tokens

showTokenCount - show count of the tokens during decoding

relativeWordBeamWidth - relative pruning beam for lookahead

growSkipInterval - interval to skip growth step

wantEntryPruning - entry pruning

Method Detail

newProperties

public void newProperties(PropertySheet ps)
                   throws PropertyException

Description copied from interface: Configurable

This method is called when this configurable component needs to be reconfigured.

Specified by:

newProperties in interface Configurable

Overrides:

newProperties in class TokenSearchManager

Parameters:

ps - a property sheet holding the new data

Throws:

PropertyException - if there is a problem with the properties.

startRecognition

public void startRecognition()

Called at the start of recognition. Gets the search manager ready to recognize

recognize

public Result recognize(int nFrames)

Performs the recognition for the given number of frames.

Parameters:

nFrames - the number of frames to recognize

Returns:

the current result or null if there is no Result (due to the lack of frames to recognize)

undoLastGrowStep

protected ActiveList undoLastGrowStep()

Because the growBranches() is called although no data is left after the last speech frame, the ordering of the active-list might depend on the transition probabilities and (penalty-scores) only. Therefore we need to undo the last grow-step up to final states or the last emitting state in order to fix the list.

Returns:

newly created list

stopRecognition

public void stopRecognition()

Terminates a recognition

recognize

protected boolean recognize()

Performs recognition for one frame. Returns true if recognition has been completed.

Returns:

true if recognition is completed.

localStart

protected void localStart()

Gets the initial grammar node from the linguist and creates a GrammarNodeToken

localStop

protected void localStop()

Local cleanup for this search manager

growBranches

protected void growBranches()

Goes through the active list of tokens and expands each token, finding the set of successor tokens until all the successor tokens are emitting tokens.

scoreTokens

protected boolean scoreTokens()

Calculate the acoustic scores for the active list. The active list should contain only emitting tokens.

Returns:

true if there are more frames to score, otherwise, false

pruneBranches

protected void pruneBranches()

Removes unpromising branches from the active list

getBestToken

protected Token getBestToken(SearchState state)

Gets the best token for this state

Parameters:

state - the state of interest

Returns:

the best token

setBestToken

protected Token setBestToken(Token token,
                             SearchState state)

Sets the best token for a given state

Parameters:

token - the best token

state - the state

Returns:

the previous best token for the given state, or null if no previous best token

getActiveList

public ActiveList getActiveList()

collectSuccessorTokens

protected void collectSuccessorTokens(Token token)

Collects the next set of emitting tokens from a token and accumulates them in the active or result lists

Parameters:

token - the token to collect successors from

showTokenCount

protected void showTokenCount()

Counts all the tokens in the active list (and displays them). This is an expensive operation.

getBestTokenMap

protected java.util.Map<SearchState,Token> getBestTokenMap()

Returns the best token map.

Returns:

the best token map

setBestTokenMap

protected void setBestTokenMap(java.util.Map<SearchState,Token> bestTokenMap)

Sets the best token Map.

Parameters:

bestTokenMap - the new best token Map

getResultList

public java.util.List<Token> getResultList()

Returns the result list.

Returns:

the result list

getCurrentFrameNumber

public int getCurrentFrameNumber()

Returns the current frame number.

Returns:

the current frame number

getGrowTimer

public Timer getGrowTimer()

Returns the Timer for growing.

Returns:

the Timer for growing

getTokensCreated

public StatisticsVariable getTokensCreated()

Returns the tokensCreated StatisticsVariable.

Returns:

the tokensCreated StatisticsVariable.

allocate

public void allocate()

Description copied from interface: SearchManager

Allocates the resources necessary for this search. This should be called once before an recognitions are performed

deallocate

public void deallocate()

Description copied from interface: SearchManager

Deallocates resources necessary for this search. This should be called once after all recognitions are completed at the search manager is no longer needed.

toString

public java.lang.String toString()

Overrides:

toString in class java.lang.Object